OpenAI’s Pando Git Backend Takes Over Its Main Code Repository

OpenAI’s in-house Git backend, Pando, has taken over the company’s entire main code repository. According to ChainCatcher, the system was developed as Codex enabled engineers to deploy more coding agents, increasing the volume of branch, push, merge, and pull-request activity. OpenAI’s previous systems frequently stalled and failed to merge changes, with reliability falling below 90%. Pando entered production after 10 weeks of development, led by one engineer responsible for its core architecture and supported by teams of Codex agents. The backend processes code pushes from different branches in parallel instead of making them wait in a queue. For code retrieval, it reuses existing Git data packets and distributes them through a cache rather than generating a separate packet for each client. After the changes, median push latency fell by nearly 80%, the system distributes nearly 1 petabyte of data daily, and its cache hit rate exceeds 99%. However, different branches are no longer guaranteed to show the same state at the same time. Unlike Cursor’s Origin, which remains Git-based and is aimed at replacing GitHub as a code-hosting service, or Anthropic’s Scion, which drops Git and moves commits, merges, and history queries to the server, Pando retains the Git client and protocol while rebuilding the backend. Engineers and agents can therefore largely keep their existing workflows.